Problem with configure script and compiler drivers

[andrea marini]

FC is the serial F90 compiler, PFC is the parallel one. F77 is the serial used for fortan 77 sources.

[leoteo]

Dear Andrea and Myrta,

thank you so much for your hints! I didn't realize that by FCFLAGS= I was inviting the configure to guess it.
So, apart from the solution in my second post, another possibility would be to specify some (non-empty) dummy FCFLAGS that do nothing but prevent the configure from guessing. The actual compilation using the compiler driver scripts would then anyhow use the options set by the compiler driver.

I have tried this and it seems to work in some cases, but there are still some strange things that I haven't figured out yet. Once they are figured out, I will post the script that I used.

Best regards,
Leopold

[leoteo]

Hi again,

so this is the script I came up with for using the compiler wrapper scripts directly

Code: Select all

#!/bin/bash
module load PrgEnv-intel fftw

export CC=cc
export CFLAGS='-g -O2 -static'

export FC=ftn
export FCFLAGS='-assume bscc -O3 -ip -static'
export UFLAGS=' -assume bscc -O0 -static'

export CPP='cc -E'
export CPPFLAGS='-ansi'

MKL="-Wl,--start-group ${MKLROOT}/lib/intel64/libmkl_intel_lp64.a ${MKLROOT}/lib/intel64/libmkl_core.a ${MKLROOT}/lib/intel64/libmkl_sequential.a -Wl,--end-group -lpthread -lm"

./configure \
--with-blas="$MKL" \
--with-lapack="$MKL" \
--with-blacs="${MKLROOT}/lib/intel64/libmkl_blacs_intelmpi_lp64.a" \
--with-scalapack="${MKLROOT}/lib/intel64/libmkl_scalapack_lp64.a " \
--with-fftw="${MKLROOT}/interfaces/fftw3xf/" \
--with-fftw-lib="$MKL" \
--with-iotk="/project/s148/tal/espresso-5.0.3-rosa/iotk" \
 --with-p2y="5.0.3"

which produces the summary

Code: Select all

#
# [VER] 3.4.1 r.3187
#
# [SYS] linux@x86_64
# [SRC] /project/s148/tal/yambo/devel
# [BIN] /project/s148/tal/yambo/devel/bin
# [FFT] External Fast Fourier transform
#
# [ ] Double precision
# [X] Redundant compilation
# [X] MPI
# [ ] OpenMP
# [X] PW (5.0) support
# [ ] ETSF I/O support
# [X] SCALAPACK
# [   ] NETCDF/HDF5/Large Files
# [  X] Built-in BLAS/LAPACK/LOCAL
#
# [ CPP ] cc -E -ansi
# [  C  ] cc -g -O2 -static -D_C_US -D_FORTRAN_US
# [MPICC] cc -g -O2 -static -D_C_US -D_FORTRAN_US
# [ F90 ] ftn -assume bscc -O3 -ip -static
# [MPIF ] ftn -assume bscc -O3 -ip -static
# [ F77 ] ftn -assume bscc -O3 -ip -static
# [Cmain] -Mnomain
# [NoOpt]  -assume bscc -O0 -static
#
# [ MAKE ] make
# [EDITOR] vim
#

In the end, however, I have to say that I liked the solution from my second post better, since it was also using the compiler drivers, but at the same time allowed yambo to do its guessing on what would be good options for the Intel compiler (which I basically copied now).
So, below is the script I used previously together with the ifort/icc scripts described in the second post.

Code: Select all

module load PrgEnv-intel fftw

export FC='/project/s148/tal/yambo/devel/ifort'
export CC='/project/s148/tal/yambo/devel/icc'
export CPP='/project/s148/tal/yambo/devel/icc -E'

MKL="-Wl,--start-group ${MKLROOT}/lib/intel64/libmkl_intel_lp64.a ${MKLROOT}/lib/intel64/libmkl_core.a ${MKLROOT}/lib/intel64/libmkl_sequential.a -Wl,--end-group -lpthread -lm"

./configure \
--with-blas="$MKL" \
--with-lapack="$MKL" \
--with-blacs="${MKLROOT}/lib/intel64/libmkl_blacs_intelmpi_lp64.a" \
--with-scalapack="${MKLROOT}/lib/intel64/libmkl_scalapack_lp64.a " \
--with-fftw="${MKLROOT}/interfaces/fftw3xf/" \
--with-fftw-lib="$MKL" \
--with-iotk="/project/s148/tal/espresso-5.0.3-rosa/iotk" \
 --with-p2y="5.0.3"

And just for reference, when I call ftn, this is what actually happens:

Code: Select all

/apps/rosa/intel/13.1.0/bin/intel64/ifort -msse3 -L/opt/cray/udreg/2.3.1-1.0400.4264.3.1.gem/lib64 -L/opt/cray/ugni/2.3-1.0400.4374.4.88.gem/lib64 -L/opt/cray/pmi/4.0.0-1.0000.9282.69.4.gem/lib64 -L/opt/cray/dmapp/3.2.1-1.0400.4255.2.159.gem/lib64 -L/opt/cray/xpmem/0.1-2.0400.31280.3.1.gem/lib64 -Wl,-L/opt/cray/atp/1.6.0/lib/ -Wl,--undefined=_ATP_Data_Globals -Wl,--undefined=__atpHandlerInstall -Wl,-lAtpSigHCommData -Wl,-lAtpSigHandler -I/opt/cray/udreg/2.3.1-1.0400.4264.3.1.gem/include -I/opt/cray/ugni/2.3-1.0400.4374.4.88.gem/include -I/opt/cray/pmi/4.0.0-1.0000.9282.69.4.gem/include -I/opt/cray/dmapp/3.2.1-1.0400.4255.2.159.gem/include -I/opt/cray/gni-headers/2.1-1.0400.4351.3.1.gem/include -I/opt/cray/xpmem/0.1-2.0400.31280.3.1.gem/include -D__CRAYXE -D__CRAYXT_COMPUTE_LINUX_TARGET -D__TARGET_LINUX__ -I/opt/cray/mpt/5.6.0/gni/mpich2-/include -I/opt/cray/libsci/12.0.00//120/interlagos/include -I/opt/fftw/3.3.0.1/interlagos/include -I/usr/include/alps -L/opt/cray/mpt/5.6.0/gni/mpich2-/lib -Wl,-rpath=/opt/cray/mpt/5.6.0/gni/mpich2-/lib -L/opt/cray/libsci/12.0.00//120/interlagos/lib -L/opt/fftw/3.3.0.1/interlagos/lib -L/usr/lib/alps -Wl,--start-group -lscicpp_intel -lsci_intel_mp -lstdc++ -lfftw3_mpi -lfftw3f_mpi -lfftw3_threads -lfftw3f_threads -lfftw3 -lfftw3f -lmpichf90 -lmpich -lpmi -lalpslli -lalpsutil -lpthread -limf -lifport -lifcore -lm -Wl,--end-group -liomp5 -lpthread

Best,
Leopold

P.S. Thank you again Myrta for providing your settings using the gnu compiler together with CRAY's libsci. I will test this as well and compare performance.

[huebener]

Just for future reference: I used the first script in the last post by Leopold to compile yambo on the ARCHER supercomputer (CRAY XC30). It worked basically fine, but I had to make a small tweak that could probably be avoided and might very well be due to my impatience/stupidity: The compiler (/opt/intel/composer_xe_2013_sp1.1.106/bin/intel64/ifort) did not recognize the option -Mnomain. So I changed in the configure script all the lines containing

FCMFLAG="-Mnomain"

to

FCMFLAG="-nofor_main"

I am sure there is a more elegant way, but this worked.

Cheers,
Hannes

[Daniele Varsano]

Dear Hannes,
many thanks.

Daniele